Skip to content
NTPU AI Learn

DS 105 · Module 2

AI Financial Intelligence with Big Data

Core question: how do we use AI to extract useful signals without confusing patterns with truth?

Start Section 1
  • Duration60 min
  • LevelFoundation
  • PrerequisiteModule 1 recommended
0 / 6 completed
Sync my progressJoin a session to keep progress across devices and share it with your instructor

Info

After you join, your completed lesson numbers are sent to your instructor together with your learner code, so they can see your progress. Use the codes your instructor gave you, and do not enter your name, company secrets or other personal data. Progress is kept with the rest of that session's data.

Learning objectives

  • Map decision questions to appropriate data types
  • Distinguish data from evidence
  • Use AI to compare multi-source records
  • Identify missing evidence
  • Audit an overconfident AI recommendation

Data ≠ Evidence ≠ Decision

8 min

Data ≠ Evidence ≠ Decision

Learning objective: map decision questions to appropriate data types, and understand that more data does not automatically produce a better decision.

The chain from data to decision
  1. DATA
  2. context + relevance → EVIDENCE
  3. analysis + uncertainty → JUDGMENT
  4. accountability → DECISION

Warning

Key concept: more data does not automatically produce a better decision : data must first be turned into evidence through context and relevance, then into judgment through analysis and uncertainty assessment, and finally into a decision a human is accountable for.
Check yourself

When does data become useful evidence for a decision?

When does data become useful evidence for a decision?

Key takeaway: Data, evidence, judgment and decision are four distinct stages, each requiring additional context and accountability to move forward.

Financial Data Universe

12 min

Financial Data Universe: Five Evidence Families

Learning objective: learn the five evidence families and practice choosing the most relevant one for a decision question.

Five evidence families
FamilyExamples
Traditional financial dataFinancial statements, price, volume, interest rates
Behavioral / transaction dataPayments, usage, customer activity
Alternative dataNews, search trends, reviews, satellite/geospatial indicators
Digital-finance dataOn-chain transactions, smart-contract events
Sustainability dataEmissions, climate exposure, ESG metrics, transition plans
Exercise: mapping decision to data family

Scenario: assessing a company's short-term repayment stress. The most relevant family is behavioral/transaction data (cash flow, payment behavior) combined with traditional financial data (debt structure) : not news sentiment or on-chain data.

  • For every decision question, first ask "which evidence family does this question depend on most?"
  • Evidence families can complement each other but cannot substitute for each other
  • Choosing the wrong evidence family cannot be fixed by more AI analysis
Check yourself

Which is the best example of alternative data?

Which is the best example of alternative data?

Key takeaway: Each evidence family has its own use; decision quality depends on choosing the family most relevant to the decision question.

Source Fitness

10 min

Source Fitness: Is This Evidence Fit for This Decision?

Learning objective: learn a lightweight evaluation matrix for whether evidence is "fit for this decision," not whether data is "technically valid."

  • Authority
  • Relevance
  • Coverage
  • Timeliness
  • Definition
  • Comparability
  • Traceability
  • Limitations

Info

Core distinction: DS 102 asks "Is the dataset technically valid?" DS 105 asks "Is this evidence fit for this decision?" This course does not duplicate DS 102's cleaning procedures.
Check yourself

Which factor is LEAST appropriate when judging source fitness?

Which factor is LEAST appropriate when judging source fitness?

Key takeaway: Judging source fitness asks "is this suitable for this decision," not "is this technically clean" or "does this match what I expected."

AI as Evidence Analyst

15 min

AI as Evidence Analyst

Learning objective: use AI to compare multi-source records, practicing requests that surface facts, patterns, competing explanations, missing evidence, uncertainty and what needs human review.

DS105_Financial_Evidence.csv (simulated data, excerpt)
company,revenue_growth_pct,debt_band,carbon_change_pct,esg_news_signal,controversy_count,source_date
A,12,LOW,-15,POSITIVE,0,2026-08-31
B,20,HIGH,8,MIXED,2,2026-08-31
C,7,MEDIUM,-25,POSITIVE,1,2026-08-31
D,15,MEDIUM,-3,NEGATIVE,3,2026-08-31
  1. Facts
    Ask AI to state only what the data shows.
  2. Patterns
    Find differences and similarities across companies.
  3. Competing explanations
    Propose at least two possible causes for the same pattern.
  4. Missing evidence
    Identify what data is still needed to judge.
  5. Uncertainty
    Flag which conclusions carry lower confidence.
  6. What would require human review
    List items that must be judged by a human, not AI output alone.
Check yourself

Which instruction best uses AI as an evidence analyst?

Which instruction best uses AI as an evidence analyst?

Key takeaway: AI should be asked to separate facts, patterns, missing evidence and uncertainty, rather than hand over a single conclusion.

AI Financial Claim Audit

10 min

AI Financial Claim Audit

Learning objective: audit an overconfident AI output, practicing that "sounds plausible" is not the same as "sufficiently supported."

Case: an overconfident AI recommendation

"Company A is clearly the best sustainable investment because it reduced emissions and received positive ESG news."

  • Claim: Company A is the best sustainable investment
  • Supporting evidence: falling emissions, positive ESG news signal
  • Missing evidence: financial stability, debt level, controversy count, data time range
  • Assumption: the ESG news signal represents overall sustainability performance
  • Alternative explanation: the emissions drop may be a short-term fluctuation, not structural improvement
  • Decision risk: investing on this alone could miss financial or governance risk

Warning

Core teaching: a recommendation can be plausible and still be insufficiently supported. The two are not mutually exclusive.
Check yourself

AI says Company A is the best sustainable investment because emissions fell. What is the strongest critique?

AI says Company A is the best sustainable investment because emissions fell. What is the strongest critique?

Key takeaway: Auditing AI output means checking the claim, supporting evidence, missing evidence, assumptions, alternative explanations and decision risk : not just the tone of confidence.

Core Exercise: AI Financial Evidence Check

5 min

Core Exercise: AI Financial Evidence Check

Learning objective: complete this module's core exercise, consolidating source fitness, facts, inference and risk.

  • Decision Question
  • Data Used
  • Source Fitness
  • Facts
  • AI Inferences
  • Missing Evidence
  • Unsupported Claims
  • Risk if Wrong
  • Human Review Required

Done

Completing this exercise counts as Core Exercise 2 (ds105-m2-ai-financial-evidence-check).

Key takeaway: Module summary: AI should expand evidence inspection, not replace evidence judgment.

Save as PDF

What next?